Avaliação de heurísticas de escalonamento de aplicações bag-of-tasks em grids computacionais adaptativas à disponibilidade de informação.

Abstract

Scheduling application on computational grids is a difficult task. This is due to high heterogeneity, large distribution and the dynamic nature of the grid environment. In order to map tasks onto resources in a efficient way, grid schedulers apply scheduling heuristics. The existing scheduling heuristics can be classified in two approaches: i) bin-packing schedulers and ii) replication schedulers. The first approach requires complete and accurate information about the applications (e.g. estimated execution time for all tasks that comprise the application) and the grid environment (e.g. current load and speed of all processors in the grid) in order to make scheduling decisions. This approach is clearly unsuitable to distributed environments like a grid in which there is no central administrative control and therefore, it is not possible to guarantee that this information will always be available or, when available, will be correct. Because of this, the second approach does not use any information. Instead, it applies the principle of task replication, achieving performance comparable to those heuristics that are based on information. On the other hand, its good performance is achieved at the expense of extra consumption of resources due to replication. In this work, we investigate these two scheduling approach and propose a new approach that takes advantage of each of them. The proposed approach consists of an adaptive solution to the grid environment according to information availability. This new approach provides a more efficient scheduling process since it reduces the grid resource wasting, without compromising the application performance.